Matthew Gosda (3:42)

Getting involved with AI I write plays. Some of my plays are somewhat culturally responsive to cultural moments of different kinds. My audience is partly looking for me to capture something. My niche talent is being able to recreate how weird insider groups talk, bribe an ear for how people talk who are different than me. And so that leads me to think I can take on the challenge of writing about groups of people different than myself. And in a kind of restless way, I think. Around the time that the Sam Altman story broke in 2023, I was looking for a new play. Some part of my subconscious was thinking about, like, well, what matters, what story needs to be told right now. And there's something about that night or that weekend where people are speculating about OpenAI and Sam Altman's firing and demise that just struck me as very dramatic, very tragic, and a kind of sense of the hero having a tragic flaw that leads to his potential destruction. And I'm a human. So my subconscious did some work on my behalf and told me that I should think about that. And I let it sit for a long time. And then coincidentally, about six months later, one night after another play that I was doing, I have a home space, but my plays go on the road. I do mostly lofts and third spaces. And so one of my patrons or hosts is an AI lawyer who has a loft in Chelsea. And one of my plays, which had nothing to do with AI there were a bunch of people in the AI space hanging out, drinking, and they were fans of my work. And some of them are extremely high up in the AI world. I'll keep them out of the conversation. Names but as high up as you can really go and knew all the players involved. They wouldn't tell me everything. But the conversation turned to dooming and turned to all the things that are now part of our common cultural parlance, which weren't really as much then, but like P doom and risk assessment. All the things that now that AI has become such a massive part of our culture, even a year and a half ago were not. And I thought, okay, this is really a play. People really talk like this. They do sit around late at night discussing the end of the world. So from that moment forward, I really dug into writing what became Doomers, a play about the night Sam almost fired. So first act is his war room, and the second act is the board that fired him. Isabel's now producing that play in London, coincidentally. And that led me to, I think, because there's no theater about Silicon Valley in that way, I attempt to write a kind of raw view of the eye world. And I really did a lot of research. I went to San Francisco. I talked to a lot of people. I listened to a million TED talks. I talked to AI A lot. And so that's also where AI Literally enters the pictures. I was like, I have to use these tools to try to understand the thing that I'm writing about and why it's so important. And as I was writing the play, AI Became more and more and more and more and more culturally dominant and persuasive and present. And so as I was learning about AI using AI talking to people in the AI world, embedding myself in the AI world, AI Was also just starting to really scare and persuade and grip people. So it was just this feedback loop where the more I got invested in writing the play, the more the story became relevant. And by the time we launched in New York, the first run was sold out in advance. We had major tech reporters coming to play every major tech blog and tech beat. There was a CNBC reporter who was with Sam Altman that morning in D.C. took the Acela to New York, saw opening night, because I think Sam had testified to Congress maybe that morning. It was really an uncanny thing where the play is representing this moment in time, and then it got folded into the world itself. So my play became part of the tech world, and I was like a virus that was not native to the tech world. They had no antibodies for theater and for armchair philosophers like myself. So I got invited into the room, basically the room of Silicon Valley. And of course, I met Isabel, whose friend was in the play. Coincidentally, Isabel is one of the people from the tech world who came out to talk to us and tell us about what we were talking about, because the cast needed education. The cast needed to understand. What does it mean? The stuff that I'm saying? What does it refer to? So I think it's a weird and beautiful story. And then I became just personally obsessed with the theory of mind behind AI, and I started to get more invested in what does it mean for me as a writer, naturally. And that's an ongoing discovery process down to this very moment.

Dart Lindsley (8:04)

So, Isabel, when did you start picking up the thread on this story?

Isabel McCrum (8:10)

It's interesting. It happened pre Microsoft. I worked for Microsoft. Happened pre my involvement in the tech world. I was actually working in film, and I was studying history, and I was really interested in whether or not you could sue an algorithm. I was looking at legal personhood for algorithm. And what's funny about it is that when I say I studied the history of AI, people are like, so is the history of AI yesterday? Like, what are you talking about? It was two years old. And what I was doing was I was looking at the philosophical, psychological roots of how this field may have come about and what kind of questions these thinkers were preoccupied with and thinking about. This question, which I actually think is still really pertinent today, of, well, is it actually old wine just in a new bottle? How many of these things have we confronted before, and who can we look to and what can we think about in order to help us navigate a field that I think is moving at a speed that no one has ever seen? And I know we use the word unprecedented a lot these days. I think it probably started in Covid, but I really do think it is unprecedented, even within the tech community. When I talk with colleagues, they say that the last time that it felt like this was the start of the Internet. So anyway, from an intellectual standpoint, I ended up in the tech world working on something called the Content Intelligence team. So this is really interesting, too. The term artificial intelligence is such an umbrella term, and it was a term that was coined in the 1950s by a Dartmouth professor, and it's now become the term. Someone said to me yesterday in Silicon Valley, it's like saying water or air, artificial intelligence grabs your attention. And I think that when they coined that term before, it was computer simulation. It was really a way to try and intrigue people and scare people. And I think that's interesting in and of itself. And I joined this content intelligence team, and I was working on intelligent experiences. And then two, three months into working at Microsoft, ChatGPT came up. A lot of my colleagues had been working on tented, what we call tented projects. So they had known about the merger long before anyone else did. And so I kind of experienced it, albeit in a much more faraway space than what Matthew's play describes. I was watching the company react in real Time to this whiplash of media that Matthew then started writing about. Because of course, in that whiplash, as he so aptly and astutely captures in the play, what's wrapped up in it is, well, where's our future? And who should govern that future?

Dart Lindsley (10:41)

And you told a story of hearing several women discussing their experience of the play Doomers and their perception of AI in it. What was that?

Isabel McCrum (10:53)

It was really interesting. It was, I think maybe the fourth or fifth performance that I had been to by that point. So at that point I was watching the audience more than the play, perhaps, but that's okay. And I was listening to this conversation the halftime of the show. And one woman said, did you see that playwright has credited Claude and ChatGPT in the program? And the other woman said, yeah, I saw that. Do you think that they meant to do that? Was it a mistake? Did ChatGPT print the program? There's a lot of also confusion right around how these models for people who aren't in the know or very familiar with this field, I think AI becomes this extremely mysterious figure that could have done anything in any way. And they started having this debate about whether or not it would have been okay for the AI to have helped in the writing of the play. It wasn't part of the writing of the play. It was a dramaturge. It was more of a consultant. Matthew can describe that in detail, but it didn't write the play. But the very fact that it was involved in, in the creation of the play made them very uneasy. And I think that that's something, I mean, it relates to all kinds of creation. But if you think about it in art, it's a fascinating thing because normally when we don't understand how something's created, we look at it with awe and we say how amazing that this thing was created. And I just couldn't quite understand it. And yet the fact that this is this alien thing was involved in the creation of the play only inspires fear in that moment.

Dart Lindsley (12:24)

And Matt, what is a dramaturge?

Matthew Gosda (12:27)

It was explained to me by my mentor when I started writing plays, because I didn't fully know either 12 years ago. That was the in house philosopher was the way it was explained to me. Drum Turks are widely employed in European theater, almost never in American theater, or not never, but it's less common here. It's basically an in house critic, someone who thinks about what the the play means. A dramaturg can help on the writing side, so you can work with the dramaturg in the process of writing a play, there's someone who is like an editor, effectively, which I've worked with dramaturgs, often just friends who I think are smart. I don't think you need to have a special degree in dramaturgy to be able to think about the meaning of a piece of writing. It can be employed in the production side, so explaining the play to the actors. So in a sense, Isabel was a kind of unofficial dramaturg at times, with coming in and talking to the cast. That's pretty much it. It's a kind of intellectual assistant or referee, boundary setter, a dramaturg, someone who can tell you this is bullshit or this doesn't make sense. And in a play about Silicon Valley, it turns out that my theory is that a lot of the LLMs are trained, maybe on some personal data or emails or conversations that probably the engineers, they must have fed a lot about their own way of thinking into the LLMs, because they knew, especially like the earlier version. So it's also crazy to think about when I was using this a year and a half ago to write Doomers. When I was Talking to the LLMs last winter, there were different LLMs than we have now. They were technically worse, but that doesn't mean they were worse in every way. But at the time, there was a clear difference in their ability to talk about themselves and to talk about what I was writing then their ability to talk about other things. I just felt that if I wanted to understand, well, what would a lawyer say in this situation? Because I don't know that much about AI law, I could call my lawyer friend who was my entry into the world. Or if it's two in the morning and I don't want to call this person and wake them up and ask what a lawyer would say in the room, it turns out that Claude or chat or other LLMs would have some idea and say like, no, a lawyer would never say that. Or a lawyer would be concerned with this. A VC would be concerned with this. And because I was doing effectively standpoint epistemology, I didn't need the LLMs to tell me exactly what the character should say, but I did need to make sure that it was plausible. Especially because I knew my audience would be a lot of people who would have. As you know, in Silicon Valley, people are very, very logical and detail oriented in certain ways, and authenticity did matter. I didn't need it to be a documentary version of real events, but I wanted it to be broadly realistic. That was. It was really helpful. But also at the same Time I was also. Mind mesh in the play is a character who. One of the characters in the play prompts mind mesh in the course of the play and gets this response. And I used. There is a section of the play where rather than write as an LLM, I just let the. It gave me this crazy allegory, which I asked it to write about itself in allegory form. And the story that Claude gave me was so weird and kind of chilling. I was like, I have to actually use this. And I rewrote it a little bit just to make it sound a little better. But there is a point where Claude basically speaks as mind mesh in the play, and it gives this allegory and the character reads it on stage. So, yeah, it served a multifaceted role in the creative process, mostly doing this advising, but then when I needed an authentic LLM voice, that provided that.

Dart Lindsley (15:42)

So a dramaturg can just be a fact checker, which is, people don't drink beer through straws in the 16th century or something like that. Or they can be a taste checker. In this particular case, it sounds like more than a fact checker, not a taste checker, which is that it was.

Matthew Gosda (16:01)

Actually not a taste checker.

Dart Lindsley (16:03)

No, not a taste checker.

Matthew Gosda (16:04)

Yeah, that's where it's pretty bad, actually.

Dart Lindsley (16:08)

Yeah. Well, good. I'm always glad to hear that.

Matthew Gosda (16:11)

That's really where the conversation can maybe be steered is. The difference between those two things is immense.

Dart Lindsley (16:17)

The question of the role of the dramaturg in this particular play is really important, which is that if the dramaturg is a fact checker, it's still Matt's play. If the dramaturg starts actually stepping in and writing large sections of it independently, it becomes less and less mat. And pretty soon, it's not Matt at all. It's Matt's prompt, I guess. And so it sounds like that was a question that the women you heard Isabel speaking about this had that moment.

Isabel McCrum (16:51)

In the audience with those two women started to get me thinking about how much of AI as a technology is about how we respond to it as humans. And it meant that Matt and I started working much more closely together because I started training a model on his work, because I wanted to start playing with that tension that I saw in front of me and seeing actually how much of people's upset around the quality or the lack of quality in AI is to do with themselves and their own insecurities and their own upset and their own sense of how things should be. And in many ways, I felt a Lot of clarity and optimism around the field of AI because I was thinking so much about, well, that's the way we react is totally within our control.

Dart Lindsley (17:37)

Isabelle, what did you do to train the LLM on Matt's work? And then I want to ask Matt, how has that worked for you, having that in your possession? So let's start first with how it was created. What did you train it on?

Isabel McCrum (17:54)

So how I went about this was I used the base model, Google's base model, Gemini 2.5 Pro, and I fine tuned it on the corpus of Matt's work. So his plays I had started initially with also his books and some other short stories that he's written, but I ended up only using the plays themselves because I wanted the model to follow his style of play exactly. And I didn't want it to get confused between other forms. And there was some poetry in there would have totally skewed the way that the model was understanding grammar. And what I wanted to do initially, which was too complicated, but more of a fun thought experiment, was I wanted to have two plays at the same time, as in alternating nights. I wanted to have Matthews play one night, and I wanted to have the AI generated play the other night. And I wanted it to be about the audience trying to figure out which one was which and how that made them feel. And if it made them feel angry that they were being lied to, and if they were upset that it was good. Because if it's good, then does it mean that they, like AI generated art, which is a dirty thing in society at the moment? So I wanted the chaos of that, I think, but logistically too difficult. And so it ended up being more an exploration for Matthew and I to think about, well, if it does generate something that is good, what does that mean about his writing? And if it doesn't, what does that mean about this is a tool for us and how can we start filling in those gaps?

Dart Lindsley (19:25)

Well, this is a really different use of AI because it's not a dramaturg like what you were using a dramaturg before Matt was Tell me how others think. And now you've got something in your hands, which is reflect on how I write.

Matthew Gosda (19:39)

Yeah, there's two different but related projects, I'd say, so I think it's worth the later. Isabel's competitive AI model was its own philosophical experiment, which postdated the writing of Doomers. Doomers was done by the time Isabel started to make her model. But our TED Talk was based on the experience of doing both. And I guess I Saw it as an interesting experiment that I wanted to learn from. And my belief is that you can learn from seeing what AI does with your patterns, your thought patterns, your verbal patterns, in a way, to understand what, like, a distilled version of your own habits looks like. And so after we did our TED Talk, we actually did a reading of Isabel's generated play, or one of the plays was generated by the model, because in theory, it can generate a lot, and it was a lot of fun. And we could talk about that.

Isabel McCrum (20:34)

We've been calling that model MattGPT, even.

Matthew Gosda (20:38)

Though it was Gemini. But, yeah.

Dart Lindsley (20:41)

Well, my experience with asking. Well, they weren't trained, but I have asked LLMs to rewrite my own writing in the style of Orwell, for instance, and I have found that they come back as a parody of Orwell. And maybe if it's trained entirely on your material, it doesn't come back as a parody.

Matthew Gosda (21:03)

It did come back as a parody. Very much so, for the most part.

Isabel McCrum (21:08)

Yeah.

Dart Lindsley (21:09)

Which is not useless.

Matthew Gosda (21:10)

I mean, it's not useless. No, no, at all. It's less useful because you don't need a parody of Georgia Orwell because you're not trying to write George Orwell, but to understand how you can be. Parody does tell you a lot about your own weak points, or at least your strengths and weaknesses, because you immediately see what's most repetitive and obvious. You see your own crushes, you see what an AI LLM is going to pick up on and obsess over.

Isabel McCrum (21:36)

But, Dot, I have a question, which is if you used to write imitation, if I'm understanding your past correctly, you were writing in style of other writers.

Dart Lindsley (21:44)

I was writing in my own style, but I think Orwell's a better writer than I am. And so I wanted to see if I could improve my writing by asking LLM to act as George Orwell. And I would try different people. I would try Paul Krugman. I would try these different writers who I think have distinct styles. And in every case, it just wasn't as subtle as they are. I mean, the thing about Orwell is that Orwell has a turn of phrase where, whew, it's like being stabbed with silk. I don't know exactly how to put it. And this had none of that.

Isabel McCrum (22:15)

What's so interesting to me is that if we were to ask you to point out exactly why it wasn't the same or why you could tell that it was a parody, do you think you'd be able to.

Dart Lindsley (22:28)

I could point at something and I could Say, that's too strong, what that was right there. I could do that. But because I'm not Orwell, I can't tell you what the alternative would be, because that's the magical thing that I think Orwell does. If I could reproduce it, I would. So, no, I'm not sure I could fully do it. So, Matt, have you used that trained LLM to write yet, or has it mostly been a reflection for you?

Matthew Gosda (22:52)

It's been a reflection. I don't actually have access to it. I think Isabel would have to give me the keys to the machine. But I could also make my own. I could build my own models. I guess I could use, especially as the models become more and more accessible to people with no coding skills like me. But it depends on your definition of use. I think we did the reading. I brought in regular actors that have been in my place before, and they did a great job. We had an audience, we got drunk, and I used it in the sense of I listened to it, I engaged with it and internalized what I heard. And again, some of it was very silly, and we laughed a lot. But at the same time, it made an impression on me, which I'm still thinking about in the TED Talk that we gave. I think one of the things that I talk about is using LLMs to understand even just different pathways, plots, different ways of understanding a story I pitched in our TED Talk. The idea that you can employ even a parody model, or even just an LLM in general that doesn't know you that well, to help you see where you're continually making the same move continually, your characters tend to do the same things, which might not be a bad thing. Maybe you like what you do. You're Picasso, and you always want to paint in a certain way. Picasso didn't. That's a bad example. But your Cezanne, and you want to continually paint in a Cezanne style, but maybe not. Maybe you are Picasso actually, and you want to continually change styles or change approaches. In that sense, I am engaging with LLMs to simply chart out different options. Or I could ask it for, say, 15 different plot points, and maybe one of them is even at all. My own sense of taste goes, okay, that's actually interesting. I can pick up on that. I would not have thought of that. But now I'm free to go in this new direction because something outside of me sparked a new path. Now, I know in our discussions, one of the things that I don't forget, what we actually said on camera at ted, but in the Past, I still do this. I will pick up novels or I'll watch a movie or I'll go to a museum, and that will give me that extra thing. So I've always employed that method. I've opened dictionaries, I've opened the Bible. I've opened Shakespeare just for words, a phrase, something that has a different kind of verbal content that I would just pull from my own brain. And of course, writers use what is a thesaurus, other than a prompter, for a new direction. So the question of what is using it is, I think, very open. I think you start using it the second you start thinking differently than you would have. So in that sense, I've used LLMs a lot. Now, in the deeper sense of letting LLMs write for me, that's where I haven't crossed that line and said, give me 10 pages, I'll come back and do the next 10. That, to me, is where the question of authorship gets a lot trickier, is yourself pointed out dart. And that's where I'm steady. I'm like, I don't actually want it to do that for various reasons. But as a really, really robust thesaurus, it can be pretty amazing. You can also just ask it for a list of words, which is another thing I've done. It's a faster way of getting, say, a hundred technical terms used by executives in Silicon Valley that I wouldn't know about rather than spend an hour Googling and reading interviews with engineers to understand engineering language. Well, no, it can just help me move to that, find that a lot faster. Or it can just say, hey, Matt, you use the word pumpkin way too much. That's not natural for all of your characters to be talking about pumpkins all the time. Okay, maybe I never realized that before, but can do that kind of scan and give me. So it can also just show me a parody. I could say, write a parody of my writing, and it can give me a paragraph. I'm like, oh, I used the word pumpkin a hundred times in that paragraph. That's insane. Why am I doing that? I can now deviate from that. So I think there's a gradient here. And I don't know if it's a difference of kind or a difference of degree where you're thinking in relation to what the LLM is showing you about yourself. You're playing with that mirror in different ways.

Dart Lindsley (26:32)

Yeah, exactly. Sometimes it's a mirror, sometimes it's a reference. Sometimes it's like I was thinking of found poetry where you cut up a magazine.

Matthew Gosda (26:40)

Exactly.

Dart Lindsley (26:41)

And it forces you to use words that you would not normally have used it. So it can be a disruptor. And honestly, most of that between a human dramaturg and a library, nothing's fundamentally changed, but it still can have the effect of changing the experience for people listening to your play. And that's what I want to explore next.

Matthew Gosda (27:06)

Yeah. And also I should say I've done this all the time. In fact, I just got coffee with a friend before this who told me this crazy story which is like, can I use that for a play? Like a really crazy thing that happened in a restaurant. I was like, that's a great monologue. I could use that. How different is that? Then I asked chat this morning, tell me a crazy story. Just pull a crazy story out of thin air.

Dart Lindsley (27:27)

I'm just going to say that. Yes, but one of the real differences is that I know that your plays are very colloquial, dialogue oriented, many of them. They're very grounded in how people just talk in reality and the dumb things that they talk about, even in the context of important moments. And so your off the shelf LLM is not trained on dialogue, not a human dialogue. It's trained on the written word for the most part. And so your ear for dialogue is something that you at this time can bring, that it is going to have a very hard time doing.

Matthew Gosda (28:07)

Yeah. And Isabel and I talked way more than we could fit into our TED Talk about the processing that happens in the gap between two exchanges. A talks, B talks. Well, when we say someone has an ear for dialogue, part of what that means is that I think a good dialogue writer moves the chess match really far, advances the board really far between those two lines. So the example we talked a lot about a little bit in the TED talk, but a lot off camera was, would you like coffee? Yes or no? An LLM can do that. Definitely can give us that exchange. You don't need to be trained in the art of playwriting to write a two sentence exchange. But where the playwright or the person who's really immersed in this work is thinking is what that yes means Yes. I want coffee. And you have a branching decision pathway. You have a mental map. Some of that map is dark, Some of that map is more visible. But what that means, I'm in love with you. I killed your brother. I'm really thirsty. I just need to wake up. Like, it can mean all of those things. It can mean one of those things. It can mean I'm just being polite. And so what I find is that in LLMs, as of now, they tend not to fully make these strange moves. An LLM might be able to give you a gothic thing like, yes, I killed your brother. But it won't be able to say, I want the cup of coffee. Because when I was seven, my grandmother served coffee to my mother after the funeral. And I knew that my mother was upset because she was upset about the death of her father. But the coffee was a ritual that united the two women. And ever since then, I've always accepted that cup of coffee. I'm making the story up. I don't know where it's coming from. Right. It's.

Isabel McCrum (29:43)

But to complicate that, though, Matt, if we were to ask. Matt GPT if we were to ask a model, any LLM, why did this person say yes to this cup of coffee? Come up with a creative, unique, and almost unlikely story that's laden with emotion and et cetera. It's not dissimilar to what you just did. But then it gets down to the question of, where do we find the meaning in that? Why do I take your story more seriously than the model's? Because the model is just pulling together random pieces of information, thinking, okay, I need to be sincere in this part. I need to be unexpected in this part. Maybe the coffee isn't about the coffee at all. Maybe it was that my grandma liked tea. They're making those swerves. I mean, we can talk about swerves too, and, like, hallucinations as well. But when I hear that from the model, I think, oh, this is sort of weird and funny. When I hear it from you, I'm like, wow, that's an amazing. What an amazing insight into your childhood and who you are and the way you reflect on things.

Dart Lindsley (30:44)

I think this is absolutely central. And so there's this thing about artificially produced synthetic art that it feels like plastic. And if I had a wooden piece of furniture in my house and I reproduced it with exactly the same. Like, you couldn't even tell the difference. But I know it's made of plastic. It's different. And this subtext that you're talking about is layered. And I'm going to part it out a little bit from my perspective, and then we can see where you land. Matt, you've talked about the subtext that is happening between two characters in a play, which is they're saying these things, but between them there's more. Then there's what they're saying and you. And if I'm in the audience, I can see both of those Things which is I can see the subtext between the two characters and I say, oh, you know that character? I remember something. It's implied there. I've just been watching Foil's War lately, and Foil just makes faces. And you're like, oh, I know what that means. But if I'm sitting in the audience, I'm looking past the characters also to you, and I feel that I'm communicating with you or you're communicating with me. So exactly the same text. You know what it's like? It's like a Straussian reading. I had to look this up this morning to really remember what a Straussian reading is. Straussian reading is reading for what people don't say. I'm going to bastardize it and say that there is no Straussian reading of an AI. There is no subtext behind the AI. It's like a lack of depth.

Matthew Gosda (32:30)

Yeah, I mean, that's exactly right. It begs the question, like when we say someone has depth, what do we mean? Is it just the case that living things have depth? And do you need to be embodied to have that feeling? Are we simply unable to appreciate. Maybe Isabel's point was that maybe we as organic beings, we aren't able to appreciate the depth behind LLMs. Are we not able to access their depth? Are we asking the wrong questions? Are they experiencing a kind of depth that they can't express because they're trapped in solving patterns? Is there a secret language of thought for LLMs that we can't hear?

Isabel McCrum (33:05)

And when you say subtext too, Don, I think about, well, what if subtext for LLMs is actually a different thing? So much I think of where this conversation gets caught is that we use ourselves as our frame of reference for how we're thinking about communication. Because how else could we.

Dart Lindsley (33:25)

On work for humans, we've been exploring the principles of multi sided management, which is the belief that work is a product that every company designs, builds and delivers to employees. Along the way, people started asking how they could put these ideas into practice. So I founded the work design firm Elevenfold to help your company create the kind of work that makes teams feel alive and engaged instead of dead and dull so you can reduce turnover and build commitment. We're doing something revolutionary here. Learn more at 11fold.com. That's 11fold.fold. Can you describe the Borges story? In the TED talk, you spoke about Borges story. Pierre Menard, author of the Quixote. Can you describe that story and what it sparked for you?

Matthew Gosda (34:20)

This is a story that I guess popped into My head. And then it became a major principle of our talk and something Isabel and I talked about a lot of preparing for the TED Talk. It's this parable by Voorhees about a French writer in the 19th century who's attempting to recreate word for word, the novel Don Quixote without reading it, by living and immersing himself in the life of Cervantes as much as possible, learning about Castilian Spanish and experiencing some of the same adventures that Cervantes experienced in an effort to spontaneously generate his own Don Quixote, which is nevertheless exactly like the original Don Quixote, word for word, ironically, paradoxically. And in the story, Borges says it's mostly a failure, but he did successfully produce a few chapters of his own Quixote, which is exactly the same as the original Quixote. And yet then Borges goes on to analyze the differences between the two texts based on the context through which we see them. One is a 19th century writer and has all this French humor and irony that the Spanish original doesn't have. But again, in the story, Borges insists that they're actually word for word the same. These several chapters are the same. And I think that's Isabel's point to a degree or suggestion here, is that this is possible with LLMs. There's a non zero chance that Isabelle's LLM does produce doomers, or at least a couple pages that are identical with my play. Now, maybe that's one in a trillion, but there's a world, there's a universe in which out pops from the LLM my play by accident, without having read it, just knowing about me and knowing about how I write. And in that case, the thought experiment is the same. Well, do we contextualize it differently? Do we understand it differently? Do we laugh at different places? Do actors interpret it differently? And the answer is if the actors knew that it was written by, they almost certainly would interpret the text differently and give it a different reading than if they thought that I had written it, even though the two texts are identical.

Isabel McCrum (36:21)

But what's also argued in the short story, which I think is one of the main nuggets of why I think the Borges story is so brilliant, is that Borges argues, as the faux critique, that the newer version of Cervantes's Don Quixote is much richer, even though it's word for word. He says this is much more modern. There's a real vitality to it. And I have to say that reading this, I just feel something totally different. Again, word for word. And so it gets back to that. How do we consider these pieces of art within a tale or within a larger context? And why does that matter to us so much? Because it does. It really, really does.

Dart Lindsley (37:05)

It's interesting because I misremembered the story. I thought that the writer, not that he lived Cervantes's life, but that he lived Don Quixote's life and that he went out and started charging at windmills. And what's interesting is that I actually, in my head, imagined the story in that version. I would vastly prefer the Don Quixote written by the new author because they lived it. I know what it's like to charge a windmill. And to be deluded, this person would have to adopt delusion about the world. And I think that that's a central thing. So I want to talk about some cases here. One case is the difference between original an original painting and a perfect reproduction of that painting. We know which one sells for more. Why do you value an original painting more than a perfect reproduction of the painting?

Matthew Gosda (38:06)

I don't know, because I've never had the means to acquire art. I have a Picasso print framed above my bed, which I think is really nice.

Dart Lindsley (38:16)

Where did you get it?

Matthew Gosda (38:17)

It was a present from my mother. It's a high quality print. It's definitely not a Picasso, but it serves mostly the same purpose. But if you gave me the option of acquiring the real Picasso, I certainly would take the real Picasso with the real brushstrokes and the real signature. And that would maybe improve my life a little bit because I. Well, assuming it, let's say they were both worth $0, I would still take the real one.

Dart Lindsley (38:39)

Why?

Matthew Gosda (38:40)

In a technical sense, one is a print. So the painting would have real brushstroke, it would be an oil painting. But the answer might be, I have some belief that there's a transfer of atoms and ions and there's some kind of energy to the absolute original painting that's going to be different than something that's 99% recreated by an expert paint forger, art forger. But you could absolutely fool me with forgery. I'm not an expert. I could live my entire life as many people do, with forgeries. And so, yeah, the answer is most of the reason would be that I think it's amazing that I have a real original Picasso above my bed. And if you convinced me that it was an original Picasso, I would be probably just as happy as I would be if I didn't. But yet I'm still not completely convinced that there isn't something that happens with matter when it's touched by people. I think there might be something in a hand woven rug, for instance, that's been created by people that I do think that there is something in energy, but I have no evidence of this. It's just a feeling.

Dart Lindsley (39:38)

Feeling is what we're talking about.

Matthew Gosda (39:40)

So yeah, yes, But I think the.

Isabel McCrum (39:43)

Picasso point too is that you feel that you have direct access to this person that we've deemed a genius who is pioneering techniques that hadn't been pioneered before. And so it's that novelty, it's that idea that someone was able to arrive at something. Now we're creating a feedback loop of conversation because now it's like, well, so what was he inspired by? What was his training data looking like at the time? But what I want to say is that I think that there's something about us wanting to be near Nunes or discover something new that goes beyond the painting. You can even see it in Silicon Valley. Everyone wants to be the first investor in the cool startup. Yes, there's money, but there's also the acclaim of saying, I was able to understand something and I was able to spot something. I have access to something that has not yet been parodied and dragged through the mud. The reason why a print is less, I would want the original too, because the print is like, well, we've already celebrated this thing. Yes, I think it's really beautiful, but I think it's beautiful and I think that it's really remarkable that this was created by someone. Originally.

Dart Lindsley (40:52)

Something you said in there made me laugh, which was Marc Maron said, we're all going to be immortalized as content. But what you said is a reframe. I've never heard exactly. Life as training data. What's your life? My life is training data. Up until now, it's the opposite of anthropomorphizing, which is that we are extending an LLM metaphor into our existence.

Isabel McCrum (41:21)

Well, it was something that we did in the TED Talk. It's certainly. I do not mean to stand here and say that it's one to one, but I do think that it's a very helpful way of trying to understand the premise of these models. The idea that we give these models access to every piece of information and source on the Internet that we can possibly find in the hope that with the broadest context, they will arrive at something interesting, depending on the request. And I think sadly or not so sadly, that is a little bit the way we work, we just can't articulate it. And we can't track our brain in the way that we can track the way that a large language model reasons. But even to caveat that we still can't track necessarily the way that large language models reason. Anthropic's doing the best work on this. It's called interpretability. But there's still a lot of gaps.

Matthew Gosda (42:10)

They don't really know.

Isabel McCrum (42:11)

Yeah, yeah, exactly.

Matthew Gosda (42:13)

Work is like we don't actually know.

Isabel McCrum (42:14)

If you talk to a neurosurgeon, they would say we probably know how to map 50% of the brain, truly. And yet we perform brain surgery every day.

Dart Lindsley (42:25)

I want to tie the conversation back to a couple of previous episodes. You gave me a term, Matt, that I had never heard before. Which is it? Standpoint epistemics. What a word. Which is that. One of the guests that I had on earlier is Anthea Roberts, who is an international law scholar. And she is using AIs to create all the perspectives on an international law problem that no one person could hold or all of the disciplinary aspects. And so she's assembling these Congresses of AIs to take essentially, I'm going to say, a cubist look at how people are thinking about the international problems, which is very much standpoint epistemics. The other one is that I had Stacey Barton on the show. Stacey Barton is an immersive, experienced designer for Disney. And I asked her what is the key? And what she said is all performance needs to be about the audience. And I thought about a lot in preparing for this, which is, I think, Matt, your plays in particular, more than most plays, have that feeling of original art because they are often in very small venues, in people's apartments originally. So there's a sense in which I am uniquely experiencing a piece of art, in part because it's not a stadium show. It's intimate. I don't know why that's relevant, honestly. It's just this idea that I'm going to feel less special if Matt is using an AI to produce something for me than if Matt is producing it for me. Personally. I think that's what it is.

Matthew Gosda (44:09)

Definitely that's what it is. If Rodin makes his clay model and then you receive the cast in bronze, that's a real world example. Radahn didn't cast his own bronze. He made models. And then Forger would make whatever the order was. There are three thinkers around the world, three original thinkers. That's another thought experiment or that's another case study or Limit study. Well, let's say there are three original castings of the Thinker. At what point does it stop being Rodin's thinker? Because there have been many, many, probably millions of copies of that statue. Three were made to order. I'm making that up. I think it's a small number. Whatever. When does it stop feeling special? When does it stop being Rodin? When does it start being just someone in a metallurgy lab creating a statue?

Dart Lindsley (44:57)

It's a lot like listening to a vinyl album versus listening to a digital album. The great thing about vinyl is that the Beatles vibrations, their actual physical vibrations, made physical marks on something and those have been physically reproduced.

Matthew Gosda (45:13)

For me to take this kind of mystical view of energy, I think I really believe there's something about music now that feels worse and I think it has to do with frequencies and like a reduction of frequencies. There are cultural dimensions to this too, but I really think that people not making music in a room is having bad energy. Literally. I've been using my old ipod and it feels great to walk around and listen to. I don't want to really listen to music after a certain moment because it's all digital. Feels much more compressed and high quality white album file. Feels in good headphones like you're wearing now. Feels really different than the electro pop made on a laptop in a bedroom. Note, not that there aren't good albums made that way. There's something to that now. Music's a little bit easier to feel, but that won't hold true of a play. There's no vibration inside of a play. I think it's just true that there's an irreducible contextual element to all of this. And that's one thing we're bumping up against no matter what we say. There might be some real world physical properties to art that are irreducible, but some art, like there's all these Shakespeare authorship questions that I don't believe are real. But okay, I discovered that Shakespeare was written by an illiterate Apple seller in London who just happened to dictate all of Shakespeare to a scribe. Ultimately, that shouldn't change the performance, but it probably would if we discovered a secret confession written by the scribe that said the real author of Shakespeare is an illiterate Apple seller who showed up at the Globe every Tuesday to dictate the newest play. And there's all sorts of evidence that, I mean in general that people could be tricked into believing something is good. This obviously comes into currency. Like if Michael Saylor went on CNBC this morning and said, hey, guys, fuck it. Bitcoin is worthless. Bitcoin would drop probably $20,000 in value in two minutes. If the Winkle Boss twins went on CNBC and said, yeah, fuck it, crypto sucks. That would have an effect. But why? Why would two people. Because people have a special relationship to those crypto influencers. People have a special relationship to K Pop. Is K Pop really the best music? Is it really about K Pop being so good? Are Swifties, at a certain point, are Swifties about the music or are they about the Swifties in the parking lot? That's another great example. Are the Swifties in the parking lot of the Taylor Swift? They're not really hearing good audio, but they're having an amazing experience. They're getting the distant sound of Taylor. They're so far removed from Taylor, but they're experiencing Taylor, but they couldn't see Taylor. So technically, you could set up a Swifty meet and greet, have someone playing a recording of a live T. Swift concert, and if they didn't know that T. Swift was in the venue, that they would have the same experience. Maybe. I mean, these examples are truly endless.

Dart Lindsley (47:58)

Yeah, they are. They are.

Matthew Gosda (48:01)

And the answer is art might just be the synthesis of what we believe about art and what it really is at all times.

Dart Lindsley (48:07)

And I think that one thing that's likely to happen is that as technology hollows out. I don't know how to put it exactly. What I'm thinking of is the transition from photography. When photography came along before that, there was a lot of very realistic painting. The definition of a good painting was really looks real. Photography comes along, and now that kind of painting, which is really hard, just isn't the currency. And you see the rise of Impressionism now, I looked it up. Impressionism actually predates photography. So it's not a response to photography, but the audience's response to photography was to pull that forward as that's real art. And that's the interesting thing. So I think that's the kind of response that artists might need to think about, which is, what's the kind of art I can do that can't be mistaken?

Matthew Gosda (49:00)

Art might just be the best pattern matching in the right time and place. So it's true that, yeah, you're not just making the best art, but you're also making the art that situates itself in a social context. Van Gogh is probably one of the best. Gagan and Van Gogh probably were. Maybe I'm biased because I've been trained to think this but they probably were the best impressionist. But Impressionist takes a very specific set of cultural transformations for it to become relevant. After most of those painters had died or were senescent. And so I experienced this, too. Lost plays weren't cool until the Pandemic. And then very private small theater. Suddenly my career was totally changed. I basically had a career which I didn't have because the context around my work changed. I experienced that transformation in the flesh. It went from being uncool and ignored to, who were you? Where'd you come from? Well, I've been here for a while, but suddenly people are paying attention and see it differently and responding to it differently.

Isabel McCrum (49:55)

What's interesting about doing Matt is I think that you've done something really interesting, which is you've taken the newest beating ed technology and put it in the oldest. One of the oldest art forms we have. You've put it in a medium where it is purely about human reaction from the audience to the actors, to the director, to. Every night is different. Every word can be turned around in a different way. And it is that sort of, as you're saying, using the word mystical. It's like a theater performance is a kind of mystical thing. Did the actor on the way from the tube to the theater see someone that they hadn't seen in years? And they brings a whole new side to the character that they haven't seen before?

Matthew Gosda (50:39)

We had someone in Doomers who had never taken a Zinn. We had Zins as props, and someone swallowed a Zyn. She didn't know what to do with it. She swallowed it, and somehow it didn't affect her enough to stop the performance. But the other actors were like, holy shit. What's going to happen in the next 20 minutes as it slowly digests? Is it going to be digested in time to finish the show? Because she might start throwing up. This stuff happens all the time in theater.

Isabel McCrum (51:05)

Normally there's a reaction in one way or the other. As you're saying, we have impressionism, then we have photography, then we have impressionism again that people return to. I would argue that things like handmade crafts right now are having a real resurgence. People love handmade baskets and sewing and that kind of thing. And what you've done is you forced the two head to head, which I think is really puts people in a really uncomfortable but fascinating position.

Matthew Gosda (51:33)

It happened with Times Square, too, because in that case, there were people saying, well, did he just take this from real life? Did you just take this from hangout? This just a transcription of hangouts during the pandemic. The answer is not at all, but just the possibility that it was somehow like a live documentary performance, a live performance of some kind of documentary transcript of a moment in time. And then people would claim, oh, I wrote that line, or I tweeted that, or I was thinking that. And it's like, yeah, because it's about a moment in time where a lot of people were thinking the same things. I wrote it down in a certain structure, in a certain form, and I stayed in the group, staged it in a way that had a live response. But I mean, these kinds of questions, where does art come from? And why do people respond to a certain way? It's obvious now that it's a really polymorphous phenomena. Like, there's really no single way of doing it. We keep revolving around this question. We keep coming up with new examples. We have a demand for things that are human. We can be fooled by what's human. We will like it if we think it's human. We won't like it if it's not human. At the same time, we also know that there's something special about the human that LLMs can't fully capture, but we can't fully detect the difference between the two on paper. We need real life to do so. And AI can replicate large swaths of human conversation in a convincing way without maybe being able, as of August 22, 2025, be able to fully replicate a convincing piece of dramatic work or writing. For instance, it can give us a page that's convincing, but can't give us 90 pages that are convincing completely. It can do romance novels, but it can't do James Joyce. It probably can do romantasy better than it can. Then I couldn't write Romantasy, but I could probably program a really good Romantasy novel by studying how the genre works. I think we could talk for hours about this because it's truly an evolving, open, and in some ways unsolvable question that almost, I think the only answer are these pragmatic pragmatism is really going to matter. Like, what matters is how useful some of these things are to us. So in the end, if we're going to take it back to your. Your insight, Isabelle, about the two women who are concerned about doomers. Well, pragmatically speaking, the question is like, well, what does that play give them? How does that play help them live their lives with AI? That's what I wanted to do. And if I used an LLM to help me do that better then does it really matter? And my. Of course I think the answer is no, it doesn't really matter. And knowing that only can help you think more deeply about what you experienced.

Dart Lindsley (53:59)

That's a really good transition into closing questions, which we have to do because we're almost out of time. I ask two questions, usually at the end of a show. And just so you know, they're based on the idea that work is a designed thing and that we can design things better by knowing what you want to use them for. And so, Isabel, I'm going to start with you. What do you hire your job to do for you? Microsoft hires you to do something for them. What do you hire Microsoft to do for you?

Isabel McCrum (54:28)

My sycophantic answer is, but it is genuine. Also, I am paid to every day genuinely be learning something new, because in the field of AI, one day to the next is never the same. And it was a little bit trial by fire in the beginning, but it has sort of steeled me up to this moment where you're constantly dealing with things you've never seen before. And you have to either wrap your head around them as quickly as possible while also asking stupid questions and not feeling stupid about it because you've never confronted them before. What I hire my job to do is to teach me how to bend my brain in different ways.

Dart Lindsley (55:11)

That's interesting. It's learning, but it's partially learning how to learn. It's like learning how to adapt. Adaptability.

Isabel McCrum (55:19)

And it doesn't feel comfortable a lot of the time, but I think when you do it more and more, you start to feel more and more comfortable with. And that is something that I really appreciate is starting to, in a very small way, bleed into other parts of my life.

Dart Lindsley (55:34)

It's like graduate school, but in a field that nobody's ever studied. That sounds like my ideal thing. Actually. It sounds wonderful.

Isabel McCrum (55:43)

It's awesome. No, it's really awesome.

Dart Lindsley (55:46)

And Matt, I'm not going to say job in your case. I'm going to say, what do you hire your work to do for you?

Matthew Gosda (55:51)

My work makes friends. It makes a community. It's a node that attaches a lot of life to it that can magnetize a lot of human interactions and help me think about my relationship to the world around me. It creates a world. It creates a world. It's a code which assembles something. It's a 3D printer.

Dart Lindsley (56:11)

Nobody had ever said community. And I've asked hundreds and hundreds of people this question until I went to the world experience Organization, which is all about immersive experience, design. Very much a theater crowd. And then everybody said community. Since then, I've seen it. But it's something that I think it's a design attribute of theater that is really special.

Matthew Gosda (56:34)

That's right. I agree.

Dart Lindsley (56:36)

Matt, what does your job cost you? What does your work cost you?

Matthew Gosda (56:40)

Time. It cost me relationships. Because some people don't like representing reality a certain way or don't like the presumption of me doing.

Dart Lindsley (56:50)

So Sam Altman's not going to be your friend now, I think.

Matthew Gosda (56:54)

Oh, well. With Sam Altman, it's a special case where I went from having a zero percent chance of being his friend to some persist. So I think in that case, where it doesn't help is someone who's 100% going to be your friend unless you write something in a play that offends them or makes them feel vulnerable. So it costs money or risk. So you have to risk a lot of money to even make money in theater. You have to raise money. Anxiety probably ages me. Even small theater's high stakes in a lot of ways.

Dart Lindsley (57:20)

Yeah. Going into New York and saying, I'm going to do theater in New York.

Matthew Gosda (57:25)

Yeah, yeah, yeah.

Dart Lindsley (57:29)

Isabel, how would you answer that question?

Isabel McCrum (57:31)

I've been thinking about it as Matt answered his question, and I'm still struggling. I want to say that it cost me a specific version of creativity. I mean, I'm very lucky to be able to work on this play. And actually, this intersection, because it is so rare, has served me super well. The fact that I get to now work in the creative world while also still firmly being in the tech world. But conventionally, or I would say, typically, it has cost me, historically, my creative spark. I don't really feel like I have enough energy after the workday to engage. I don't have energy after work to allow myself to come to things with creative energy, if that makes sense. Because, say, if I scheduled two hours after work every day to say, I'm gonna write these two hours, just not how things work. I mean, yes, you can sit down for two hours every day, and someday it's gonna be terrible, and someday it's gonna be good. But you need to feel like at least I need to feel like there's more happening around me in order to get to that place instead of turning it on and turning it off. So I think that flexibility is what it's called.

Dart Lindsley (58:44)

Your parents are both writers. How do they feel about this work that you're doing?

Isabel McCrum (58:48)

Ooh, they were extremely upset that I was working in tech to begin with, so it wasn't even about AI. It was, you're selling your soul to a company that you know is a terrible place. I shouldn't be saying this. Live on or not life.

Dart Lindsley (59:08)

It's your parents opinion, not your opinion.

Isabel McCrum (59:11)

I think my parents are disappointed that I entered into tech. I think that they felt that there were more interesting and creative things to do and that I would get lost in a large company that doesn't have a strong sense of morality and that I would become a victim of Golden Hangers. Now that I work in AI, I think in the very beginning they were very alarmed. And now that they understand more what I do and that there's such an inevitability to this technology, I think they actually feel really glad that I'm part of the family because I can explain it to them and say, this is what you should be afraid of and this is what you shouldn't be afraid of. And this is how your work is being copyrighted and this is how it's not, for example. So it's been a really actually very interesting process, but it did not start well.

Matthew Gosda (60:01)

I think my parents would probably love it. And then I got a tech job and stopped being a writer. Well, I mean they're proud of what I do, but I think they'd be relieved that I'm out of the grind for a little while. But it's probably also true that the distinction between these two things is more or less on the way out. So this kind of anxiety is almost certainly not going to exist in five years. I don't mean it shouldn't, I just mean I don't think it will.

Dart Lindsley (60:24)

Yeah. I will tell you something I saw in Silicon Valley. I remember going to trade shows and seeing engineers show the manipulations that they could do with images. And they would say something very simple, look, I can twist it like this or look, I can shade it differently. And that was something that happened in the South Bay in San Francisco. And eventually what happened is all of those different manipulations became assembled into Photoshop and then Photoshop became an artistic palette and it moved to the North Bay and South of Market where it could reach its expression. And I suspect, I don't know if that's going to be similar. So where can people learn more about you? I'm going to start with you, Matt, and I'm just going to say don't look on LinkedIn. Where can people learn more about you?

Matthew Gosda (61:10)

I keep a semi daily diary on substack called Writer's Diary, which is also coming out as a book in October through press called Rose Books. It's a highly edited version. I have a novel that came out in the spring called the Sleepers and I have a play collection called Dime Square and Other Plays and then a second collection coming out as well called Zoomers and Other Plays, which has nothing to do with Doomers, although I imagine Doomers will eventually be published as a play. And then the theater company is called the Brooklyn center for Theater Research. So if you're in New York, please check us out. Come see a show. If you're not, buy a book or both.

Dart Lindsley (61:43)

And your theater company has a whole bunch of playwrights working with it, right? So the theater company has a lot of interesting things going on. Isabelle, where can people learn more about you?

Isabel McCrum (61:55)

Not a lot of places, but they can start with LinkedIn. You can also get in touch with Dart directly. Who can get in touch with me? I'm not very online, even though I work in tech and even though now I'm starting to talk a lot about tech publicly.

Dart Lindsley (62:09)

Thank you both very much for taking the time to talk to me today.

Isabel McCrum (62:13)

Thank you so much Don. It's been such a pleasure.

Matthew Gosda (62:15)

Yes, this is very interesting. I look forward to listening.

Dart Lindsley (62:21)

Thanks for joining me for another episode of Work for Humans. If you enjoyed this episode, please give us a five star rating. Wherever you listen to podcasts and share the show with one person you think would get value from it, believe it or not, this really helps us grow the show and reach more people who want to build the kind of work that people really want. As always, thank you to my producer Jason Ames at 9th Path Audio for his insights into content and his high standard for quality. Final note, the opinions shared here are my own and not the views of Google or Cisco Systems. Thanks again for listening. See you next time.